reducing total power consumption method in cloud computing environments
TRANSCRIPT
-
8/2/2019 Reducing Total Power Consumption Method in Cloud Computing Environments
1/16
International Journal of Computer Networks & Communications (IJCNC) Vol.4, No.2, March 2012
DOI : 10.5121/ijcnc.2012.4205 69
Reducing Total Power Consumption Method inCloud Computing Environments
Shin-ichi Kuribayashi1
1Department of Computer and Information Science, Seikei University, Japan
E-mail: [email protected]
Abstract
The widespread use of cloud computing services is expected to increase the power consumed by ICT
equipment in cloud computing environments rapidly.
This paper first identifies the need of the collaboration among servers, the communication network and
the power network, in order to reduce the total power consumption by the entire ICT equipment in cloudcomputing environments. Fivefundamental policies for the collaboration are proposedand the algorithm
to realize each collaboration policy is outlined.Next, this paper proposes possible signaling sequences to
exchangeinformation on power consumption between network and servers, in order to realize the
proposed collaboration policy. Then, in order to reduce the power consumption by the network, this
paper proposes amethod of estimating the volume of power consumption by all network devices simply
and assigning it to an individual user.
Keywords
Reducing power consumption, collaboration, cloud computing environments
1. IntroductionGlobal warming is a top-priority issue requiring an urgent, global-scale response. There is a
high hope that the use of ICT will dramatically reduce power consumption, for example, by
optimizing the collection and delivery of goods and reducing the amount of travel[1],[2].
However, the widespread use of ICT equipment is expected to increase the power consumed by
ICT equipment rapidly and it is recognized that the power consumption of ICT equipment
themselves should be one of key issues.
Cloud computing services are rapidly gaining in popularity[3]-[7].They allow the user to
rent, only at the time when needed,only a desired amount of computing resources
(processingability and storage capacity) out of a huge mass of distributedcomputing resourceswithout worrying about the locations orinternal structures of these resources. It isanticipated
that enterprises will accelerate their migrationfrom building and owning their own systems torenting cloudcomputing services because cloud computing services are easyto use, and can
reduce both business costs and environmentalloads. The cloud computing environments
require a huge amount of ICT equipment such as servers, storage devices, communicationnetwork devices and client terminals. Therefore, it is clear that the widespread use of cloud
computing services will greatly contribute to a rapid increase on ICT power consumption.
Most conventional measures to save the power consumed by servers and the communication
network have been discussed and implemented independently [8]-[19]. For example, slowing
-
8/2/2019 Reducing Total Power Consumption Method in Cloud Computing Environments
2/16
International Journal of Computer Networks & Communications (IJCNC) Vol.4, No.2, March 2012
70
the processing of a server to reduce its power consumption can prolong not only its processing
time but also the bandwidth holding time in the network, which in turn increases the power
consumed by the network. Conversely, raising the processing speed of a server increases its
power consumption but reduces the processing time, and consequently reduces the power
consumed by the network. This may reduce the total power consumption. Therefore, it isimportant to take an integrated approach to saving the power consumed by both servers and the
network.
To reduce the power consumed by the network, it is necessary to measure the power
consumed by all network devices accurately and simply. One method proposed to estimate the
total network power consumption from the volume of shipment of network devices. It isnecessary to estimate power consumption more accurately if the emission of carbon dioxide
should be calculated from power consumption of each network user. However, it is not realistic
to seek to find carbon footprint for each packet in the same way that the transportation system
finds a carbon footprint for each package. It is necessary to consider a simpler way of
estimating the power consumed by an individual network.
Moreover, it is expected that the percentage of power generated by renewable power, such as
photovoltaic power generation and wind power will continue to rise. Since the power generated
by these natural means varies over time considerably, the electric power storage (high-capacity
batteries) needs to be introduced to stabilize power supply [19],[20]. This implies that it is
necessary in the future to consider cases where ICT equipment operates under the condition that
the total power supply available is restricted.Therefore, it is required to take the restriction of
the available power supply into consideration.
The rest of this paper is organized as follows. Section 2 explains the system model for cloudcomputing environments and identifies fundamental policies that should be adopted for the
three parties - servers, the communication network, and the power network - to collaborate with
each other to reduce the total power consumed by the entire ICT equipment in cloud computing
environments.The algorithm to realize each policy is also outlined.Section 4proposes possible
signaling sequences to exchange information on power consumption between network andservers, in order to realize the proposed policy in Section 3. Section 5 proposes a method of
estimating the volume of power consumption by all network devices simply and assigning it to
an individual user, in order to reduce the power consumption by the network.Finally, Section 6
presents the conclusions. This paper is an extension of the study inReferences [22]and [23].
2. Fundamental policies for reducing total power consumptionin cloud
computing environments
2.1 Proposed system model for cloud computing environments
Figure 1 illustrates the system model for cloud computing environments, in which there are
multiple areas at different locations and each area has a resource set of servers (processing
ability), bandwidth which connects the selected server to the client and electric power storage
which provide the power to ICT equipment in its area. Electric power storagewith limited
electric capacity is treated as the power network. Here, the power restriction to ICT equipment,
which could happen in the future, is taken into consideration.When a request occurs, one area is
selected from among k areas, and required amounts of processing ability,bandwidth and power
capacity are simultaneously allocated in the selected area.
-
8/2/2019 Reducing Total Power Consumption Method in Cloud Computing Environments
3/16
International Journal of Computer Networks & Communications (IJCNC) Vol.4, No.2, March 2012
71
2.2Fundamental policies for reducing total power consumption
2.2.1 Need of the collaboration among servers, the communication network and the
power network
Case 1 in Figure 2 shows an example in which the line speed is decreased to reduce the
power consumed by the network. While the power consumed by the network is reduced, thelower line speed increases the time it takes for servers to transfer data, making it likely that
thepower consumed by servers will increase. This may increase the total power consumption.
Conversely, case 2 in Figure 2 shows an example of reducing the power consumption of servers
by slowing its processing clock. While the power consumed by servers decreases, the
processing time increases, which can in turn increase the time during which a bandwidth in thenetwork is retained, and consequently the power consumed by the network increases.
Therefore, it is necessary to consider integrated measures involving servers and multiple
networks(Policy III
I ). For simplicity, this paper assumes asingle network.
The conditions for measures to reduce power consumption by the server or the network tobe effective areas below. Let X
1be the total power saved by a power consumption reduction
measure taken for the network, and Y1
the total additional power consumed as a result by
servers and client terminals affected by this measure.Any measure to reduce the power
consumed by the networkshould be taken only whenX1>Y1 .Similarly, let X2 be the total power
saved by a power consumption reduction measure taken for servers, and Y2
the total additional
power consumed by the network as a result. Any measure to reduce the power consumedbyservers should be taken only whenX
2> Y
2.
Figure 1. System model for cloud computing environments
Cmaxj: Maximum size of processing ability in area #jNmaxj: Maximum size of bandwidth in area #jPmaxj: Maximum capacity of electric power storage in area #j
: Servers (processing ability) : Link (bandwidth)
Network
Area #1
Cmax1
Nmax1
Cmax2
Nmax2
Cmaxk
Nmaxk
Area #2 Area #k
Pmax1 Pmax2 Pmaxk
: Electric power storage (high-capacity batteries)
-
8/2/2019 Reducing Total Power Consumption Method in Cloud Computing Environments
4/16
International Journal of Computer Networks & Communications (IJCNC) Vol.4, No.2, March 2012
72
The other possible measure is to increase the power consumption on the one side but may save
more power on the other side. For example, raising theprocessor clock frequency of the server,
in turn, reduces the bandwidth holding time and consequently reduces the power consumedbythe network (Figure 3). This could decrease the total power consumption by both servers and
the network.
2.2.2 Collaboration between servers and communication network
time
Energy consumption by server
time
Energy consumption bythe network
Increasedconsumption
Decrease bandwidth
[Case 1] Reducing energy consumptionby the network
Figure 2. Examples of issues when the collaborative reduction of energy consumption is not implemented
Decreasedconsumption
[Case 2] Reducing energy consumptionby server
Energy consumption by server
Energy consumption bythe network
decrease processing power
Decreasedconsumption
Increasedconsumption
time
time
Clientterminal A
Network
: Network devices such as router, switch or transmission system
Bandwidth allocated between end system A and end system B
Servers B
Z1
-
8/2/2019 Reducing Total Power Consumption Method in Cloud Computing Environments
5/16
International Journal of Computer Networks & Communications (IJCNC) Vol.4, No.2, March 2012
73
[1] A measure that has already been proposed to reduce power consumption includes
assembling as many servers and network devices as possible in some areas and keeping manyof them in the sleep state for a long time. However, if devices are aggregated independently,
mismatches between servers and the communication network can occur, as shown in Figure 4-
(1). As a result, the number of servers or network devices in the sleep state is reduced, or itbecomes impossible to provide services in the worst case. To avoid such a situation, it is
necessary to aggregate servers and network devices simultaneously as much as possible (Policy
I I
I II I
I I ), as shown in Figure 4-(2). This increases the number of servers and network devices that
can be put in the sleep state, resulting in a drastic reduction in the total power consumption. The
possible aggregation algorithm according to Policy II is outlinedas follows:
First, we focus on processing ability.
1) Area Xc which used the largest amount of processing ability and area Yc which used the
least amount of processing ability are selected from among m1 areas. m1 is the number of areas
which are not in sleep mode. The algorithm will be stopped if both Xc and Yc could not be
selected.
2) If all requests currently processed in area Yc could be migrated to area Xc (both processingability and bandwidth should be migrated), area Yc will get into sleep mode. Then go back to
1). Otherwise, requests will not be migrated and stop the algorithm.
Then, we focus on bandwidth, assuming that the aggregation in step 1 is not
executed.
1) Area Xb which used the largest amount of bandwidth and area Yb which used the least
amount of bandwidth are selected from among m1 areas. The algorithm will be stopped if bothXb and Yb could not be selected.
2) If all requests currently processed in area Yb could be migrated to area Xb (both processing
ability and bandwidth should be migrated), all requests currently processed in area Yb are
migrated to area Xb and area Yb gets into sleep mode. Otherwise, requests will not be migrated
and stop the algorithm.
If the final number of sleep areas in step 1 is equal or more than that in step 2, the
aggregation executed in step1 is adopted. Otherwise, the aggregation executed in step 2 isadopted.
-
8/2/2019 Reducing Total Power Consumption Method in Cloud Computing Environments
6/16
International Journal of Computer Networks & Communications (IJCNC) Vol.4, No.2, March 2012
74
Figure 5 illustrates an example of request migration according to the algorithm explained
above.There are three areas in the figure. For simplicity, the figure shows only processing
ability. First, Area 1 becomes area Xc, and area 3 becomes area Yc. Next, area 2 becomes area
Xc, and area 3 becomes area Yc.Finally, all requests is aggregated in area 3, and hence devices
in areas 1 and 2 go to sleep mode.
[2] Preventing a part of the servers or network devices in the sleep mode from going back tothe normal mode can be equivalent to saving the power that might be consumed by these
devices if they were actually put in the normal mode. Figure 6 illustrates an example. It is
assumed that server 1-1 (sleep mode) at area #1 needs to be put in the normal mode when a new
request is generated. If it is put in the normal mode, power consumption may rise significantly.
In this case, it is better that a new request be allocated to area #2 which still has a spareprocessing ability. This prevents the significant rise in power consumption. If area #2 does not
have enough processing ability to handle a new request, server 1-1 at area#1 will be put in the
normal mode.Therefore, it is necessary to take any action to keep or put servers or network
devices in the sleep mode as much as possible (PolicyIII).
2.2.3 Collaboration among servers& communication network and power network
Figure 4. Collaboration image between servers and communication network
Vacant
Vacant
Vacant
Vacant
Vacant
Vacant
Area #1 Area #2
Processingability
Bandwidth
Sleep state
Sleep state
Aggre-gation Processing
ability
Bandwidth
Processing
ability
Bandwidth
Aggre-gation
Aggre-gation
Aggre-gation
Sleep state
Aggre-gation
Aggre-gation
Aggre-gation
Aggre-gation
(1)
()
Joint multiple resource allocation model in Reference [6] is applied here.
Area #3
Area #1 Area #2 Area #3
Area #1 Area #2 Area #3
Cmax1
Nmax1
-
8/2/2019 Reducing Total Power Consumption Method in Cloud Computing Environments
7/16
International Journal of Computer Networks & Communications (IJCNC) Vol.4, No.2, March 2012
75
[1] If servers and network devices are aggregated independently without taking electricpower storage into consideration, mismatches similar to Figure 4-(1) may arise and the amount
of power available to servers and network devices may be reduced, as shown in Figure 7-(1).
As ICT equipment cannot operate without electric power, it would be one possible approach to
aggregate servers and network devices in the area which has a largest amount of available
electric power capacity (Policy IV), as shown in Figure 7-(2). This can prolong the time during
which services are provided. The possible aggregation algorithm according to policy IVis
outlined as follows:
Area S which has the largest available electric power is selected from among m 2
areas. m2 is the number of areas which are not in sleep mode.
First, we focus on processing ability.
1) Area Tc which used the least amount of processing ability is selected from among (m2-1)
areas.
2) If all requests currently processed in area Tc could be migrated to area S, all requests
currently processed in area Tc are migrated to area S (both processing ability and bandwidth
should be migrated) and area Tc gets into sleep mode Then go back to 1). Otherwise, requests
will not be migrated and stop the algorithm.
Then, we pay attention to bandwidth, assuming that the aggregation in step 2 is notexecuted.
1) Area Tb which used the least bandwidth of resource is selected from among (m2-1) areas.
2) If all requests currently processed in area Tb could be migrated to area S, all requests
currently processed in area Tb are migrated to area S (both processing ability and bandwidth
should be migrated) and area Tb gets into sleep mode. Then go back to 1). Otherwise, requests
will not be migrated and stop the algorithm.
If the final number of sleep areas in step 2 is equal or more than that in step 3, the
aggregation executed in step2 is adopted. Otherwise, the aggregation executed in step 3 is
adopted.
Area 1 Area 2 Area 3
Figure 5. Example of area migration according toPolicy II
Sleepmode
Aggregate
Aggregate
Sleepmode
Sleepmode
vacant
used
Area 1 Area 2 Area 3Area 1 Area 2 Area 3
Figure 6. Example of server selection accordingto Policy III
Normal mode
Server1-1 Server1-2
Area #1
Server2-1 Server2-2
Area #2
Sleep mode Normal modeNormal mode
New request
X
It is better for a new request to select normal-modeserver (Server 2-1) first, instead of selecting sleep-mode server (Server 1-2).
-
8/2/2019 Reducing Total Power Consumption Method in Cloud Computing Environments
8/16
International Journal of Computer Networks & Communications (IJCNC) Vol.4, No.2, March 2012
76
Although policy IVaims to reduce the total power consumption as well as policy III, the
approaches of reduction differ. That is, policy IV tries to maximize the number of requests
which can be processed while policyIII tries to maximize the number of equipment of the sleep
state. Policy IV is effective in the case where each area does not enough amount of electricpower. Therefore, it is proposed to adopt policyIIIwhen all areas have enough amount of
available electric power and to adopt policy IVwhen multiple areas don't have enough amount
of available electric power. The possible aggregation algorithm according to the policies III and
IVis outlined as follows:
1)First, we compare the proportionate size of resource (X), comparing the size of required
resource with the maximum resource size, and that of electric power (Y), comparing the size of
required electric power with the remaining electric power capacity.
2) If Y is more than X, processingability and bandwidth would be aggregated in the areawhich has a most available electric power capacity from among areas. Otherwise, processing-
ability and bandwidth would be aggregated in the area which has a least available resource
from among areas.[2] If the specific area cannot get sufficient power supply, it is impossible to provide services
even if it has enough spare processing ability and bandwidth. Therefore, when the remaining
power capacity drops below a certain level in one area, it is necessary to take some measures,
such as allocating new requests to other areas which have a spare power capacity, or restricting
any new requests (PolicyV) .
Figure 7. Collaboration image among servers, communication network and power network
Area #1 Area #2 Area #3
Vacant Vacant Vacant
Vacant Vacant
Vacant
Processing-ability& bandwidth
Area #1 Area #2 Area #3
Area #1 Area #2 Area #3
Aggre-gation
Aggre-gation
Aggre-gation
Aggre-gation
Electric powerstorage (battery)
Processing-ability& bandwidth
Processing-ability& bandwidth
(1)
()
Pmax1
Electric power
storage (battery)
Electric powerstorage (battery)
Vacant Vacant
Vacant
Vacant VacantVacant
-
8/2/2019 Reducing Total Power Consumption Method in Cloud Computing Environments
9/16
International Journal of Computer Networks & Communications (IJCNC) Vol.4, No.2, March 2012
77
3. Possible signaling sequences to exchange information on power
consumption between network and servers
Torealize Policy I in Section 2.1, it is necessary to collect information on the amount of
power that will be consumed by the network and servers respectively. The following fouralternative methods can be conceivedfor the collection of the information:
Power consumption per unit time by servers and that by the network are
uniquely specified based only on the type of requesting application (VoIP, webaccess, file
transfer etc.)
Power consumption per unit time by servers and that by the network areuniquely specifiedbased only on the bandwidth used.
Power consumption per unit time by the network is uniquely specified based
on the bandwidth used, and the number of nodes traversed between the originatingand
destination nodes. Power consumption per unit timeby servers is specified by the user at the
time ofsubscription.
Power consumption per unit time of each service request is estimated by
exchanging the information among server, client terminal and the network when a connectionisestablished.
Considering the accuracy of collected information, alternative 4 would be the best
solution.
Figure 8 proposes possible signaling sequences for alternative 4. The power consumption
of transmission equipment between nodes is included in the power consumption by the node
here. In case 1 of Figure 8, the originating client terminal A indicates the reducible volume of
power consumption (DA) which could be reduced with an action by terminal A. In addition,
terminal A indicates the value of additional bandwidth holding time (Tx), which will be added
as a result of the action by terminal A. All nodes on the route calculates the additional powerconsumption by the node (U1, U
t, U
2) based on Tx and transfers the calculation result toward
the destination server B. The destination node (node 2) indicates the total power consumption
by all nodes (U12
) on the route to server B, in addition to DA
and Tx. Server B calculates the
reducible volume of power consumption (DB) based on T
x, which could be reducedwith an
action by server B. Finally, server B judges whether the proposed action by terminal A
shouldbe taken or not as follows:
If (DA+DB)> U12 , then take the proposed action to reduce power consumption at terminal
and server, and 1 is set toflag in Connect message. Otherwise, 0 is set to flag. Server B
returns Connect message toward terminal A. Terminal A and all nodes on the route can
understand that the proposed action should be taken when flag is 1 . When flag is 0, any
action should not be taken.In case 2 of Figure 8, terminal A indicates the additional volume of power consumption
(UA which could be increased with an action by terminal A. Ty is the value of bandwidth
holding time which could be decreased as a result of the action by terminal A. The very similar
procedure discussed for case 1 is also appliedto case 2.
In case 3 of Figure 8, the originating node (node 1)indicates the power consumption volume
(D1) toward server B, which could be decreased with an action by node 1. In
-
8/2/2019 Reducing Total Power Consumption Method in Cloud Computing Environments
10/16
International Journal of Computer Networks & Communications (IJCNC) Vol.4, No.2, March 2012
78
Client terminal AOriginatingNode (Node 1)
Setup DA,Tx)
Setup
DA,U ,TxSetup
DA,U1, Ut,Tx
Connect flag
Setup
DA,U 2,Tx
Connect
flag
TransitNode
Connect flag
Connect
flag
DestinationNode (Node 2)
*
: If (DA+DB) > U12 , then take an action to reduce energy consumption at end systems and set flag=1*
OriginatingNode (Node 1)
Setup UA,Ty)
Setup
UA,D ,TySetup
UA,D1,Dt,Ty
Connect flag
Setup UA,D 2 ,Ty
Connect flag
TransitNode
Connect flag
Connect
flag
DestinationNode (Node 2)
: If (UA+UB) < D12 , then take an action to increase processing power at end systems and set flag=1*
OriginatingNode (Node 1)
Connect
UB D 2
SetupSetup
D
,VxSetup
D1,Dt ,VxSetup
D 2 ,Vx
Connect UB
TransitNode
Connect
UB D 2
DestinationNode (Node 2)
: If D12 > (UA+UB), then take an action to reduce energy consumption at networks nodes and set 1 to flag
Connect Ack(flag)
Connect
UB D 2, Vx
Connect Ack flag
Connect Ack flag
Connect Ack flag
OriginatingNode (Node 1)
Connect
DB,U 2
SetupSetup
U
,VySetup
U1,Ut ,VySetup
U 2 ,Vy
Connect
DB
TransitNode
Connect DB,U 2
DestinationNode (Node 2)
: If U12 < (DA+DB), then take an action to increase energy consumption at network nodes and set 1 to flag
Connect Ack(flag)
Connect Ack
flag
Connect
DB,U 2,Vy
Connect Ack
flag
Connect Ack
flag
flag: 1 means that an action to reduce the total energy consumption should be taken.
0 means that an action to reduce the total energy consumption should not be taken.DA: Energy consumption volume which can be decreased by terminal A per unit timeDB: Energy consumption volume which can be decreased by server B per unit timeD1 : Energy consumption volume which can be decreased by node 1 per unit timeD
: Energy consumption volume which can be decreased by transit node per unit timeD2 : Energy consumption volume which can be decreased by node 2 per unit time , D12=D1+Dt+D2,UA : Energy consumption volume which can be increased by terminal A per unit timeUB : Energy consumption volume which can be increased by server B per unit timeU1 : Energy consumption volume which can be increased by node 1 per unit timeU
: Energy consumption volume which can be increased by transit node per unit timeU2 : Energy consumption volume which can be increased by node 2 per unit time, U12=U1+Ut+U2
:places where actions to reduce total energy consumption be taken
Figure 8. Possible signaling sequences for alternative 4 to collect information on energy consumption
*
*
*
*
*
Server B
Client terminal A Server B
Client terminal A Server B
Client terminal A Server B
-
8/2/2019 Reducing Total Power Consumption Method in Cloud Computing Environments
11/16
International Journal of Computer Networks & Communications (IJCNC) Vol.4, No.2, March 2012
79
addition, node 1 indicates the value of reducible bandwidth (Vx) which could be reduced as
a result of the action by node 1. The transit node and the destination node (node 2) calculate the
reducible power consumption by the node (Dt, D2) based on Vx and transfer the calculation
result toward the destination server B. Node 2 indicates the total power consumption by all
nodes (D12) on the route to server B, in addition to Vx. End system B calculates the additionalvolume of power consumption by server B (UB) based on Vx, and judges whether the proposed
actionby node 1 should be taken or not as follows:
If D12> (UA+UB), then take the proposed action to reduce power consumption at nodes and
1 is set to flag inConnect Ack message. Otherwise, 0 is set to flag.
Terminal A returns Connect Ack message with flag toward server B. Server B and all nodes
on the route can understand that the proposed action should be taken when flag is is 1. Whenflag is 0, any action shouldnot be taken.
In case 4 of Figure 8, the originating node (node 1) indicates the power consumption volume
toward server B, which could be added with an action by node 1 (U1).Vy is the value of
additional bandwidth which could be increased as a result of the action by node 1. The very
similar procedure discussed for case 3 is alsoapplied to case 4.
Note that the information exchange proposed in the above would require complicated
processing. It would be possible to eliminate the need for such information exchange by
estimating the power consumed at server and terminal.Figures 9 and 10 show examples of the
electric power measurementof a server that performs file transfer or handles web access
respectively. If these data are stored in the network in advance, it may be possible to estimate
the power of the server by measuring the volume of data transferred, or the number of
transactions processed.
Figure 9. Data transmission rate vs. server electricpower (File transfer)
Data transmission rate [kbytes/sec] Number of transactions [/sec]
Figure 10. Number of transactions vs. server electricpower (Web access)
Serverelectricpower
[w]
Serverelectricpower
[w]
-
8/2/2019 Reducing Total Power Consumption Method in Cloud Computing Environments
12/16
International Journal of Computer Networks & Communications (IJCNC) Vol.4, No.2, March 2012
80
4. Measurement of power consumption by all network devices and its
allocation to individual user
To reduce the power consumed by the network, it is required to measure the power
consumed by all network devices accurately and simply.For transportation services, it is necessary to distribute the fuel (or power) consumed during
transportation to individual cargo owner when a car carries a mix of cargoes of multiple
owners. As an individual owner cannot make this allocation, the carrier could make it based ondata submitted by each owner about the weight of and the distance covered by its cargo. Each
cargo owner receives data on its fuel (or power) consumption from the carrier. The carrier will
collect the data on power consumption by calculating consumed fuel of each truck or attaching
a wireless tag to each package or managing carbon footprint, which is the history of where and
how much carbon dioxidehas been emitted, for each package.As the unit of cargo carried in the network may be a packet, a flow of packets or a
established connection, it is not feasible to adopt the same way practiced in transportation
system. To solve this problem, this section considers a method in which multiple probes are
installed in the network to measure traffic distribution at different hours of day, and based on
that data, the power consumed by the network is calculated and assigned to individual users.For the purpose, a new MIB (ex. Reference [19]) should be defined to collect data on power
consumption by each network device, and a system can be constructed to collect data
periodically. The information on packet routing paths in different hours of day is also
necessary. The routing path of user packets in the network can be identified by sending route
identifying packets (ex. tracert) between all edge nodes periodically. The ratio of powerconsumptionassigned to a target traffic volume is given by
[target traffic volume]*[total energy consumption by given node]------------------------------------------------------------------------------------------------------------------ (1)
[total volume of traffic that goes through a given node]
This is calculated at different hours of day.
Figure 11 illustrates an image how to allocate power consumption atnetwork node to
individual traffic flows. The entire system required for the above proposed method is illustrated
in Figure 12. The monitoring system will create the traffic distribution table and the
management table of packet routing path in different hours of day. Here, we suppose the
network system which has m servers or terminals, m edge nodes and k transit nodes. Each
server or terminal is supposed here to be accommodated to the dedicated edge node.When table
1 and table 2 are supposed to be traffic distribution table and packet routing path table
respectively, the power consumption P12assigned to traffic volume v12is calculated according to
Equation (1) as follows:
P12=Pe1*r+Pe2*r+Pt2*v12/{v12+v1m+v23+vm(m-1)} (2)
r = v12/{(v11+v12++v1m)+(v21+v31++vm1)}
-
8/2/2019 Reducing Total Power Consumption Method in Cloud Computing Environments
13/16
International Journal of Computer Networks & Communications (IJCNC) Vol.4, No.2, March 2012
81
Note that Pe1, Pe2, Pt2 are the total power consumption byedge node 1, edge node 2,
transit node t2 , respectively. As a result, the total power consumption POWg assigned to server
g (or terminal g) is given by
mPOWg= (vgi +vig) (3)
i=1
F2
F1
Node G
*Energy consumption of node G is divided to each flow(F1 and F2) based on traffic volume
Figure 11. Image of allocating energy consumption
at node G to individual user
Network
Monitoring system of trafficdistribution in the network
Data collecting system on energyconsumption by network devices
Figure 12. System for measuring and monitoring of energyconsumption by network devices
: Traffic probing equipment : MIB for energy consumption
NodeClientterminal A Server B
NodeTransitNode
Sourceedge node
Destination edge node
Node 1
Node 2
Node 1
v21
v11
Node m
Node 2
v22
v12
Node m
v2m
v1m
vm1 vm2 vmm
vij: Traffic volume transmitted from edge node i towardedge node j
Table 1. Example of traffic distribution table
Edge nodepair
Edge or transit node
Pair1->2
Node 1
Node 2 Node m
Pair i->j : Pair of source edge node i and destination edge node j0: Not on the routing path, 1: On the routing path
Pair1->3
Edge node
Node t1 Node t2 Node tk
Transit node
1 1 0 10 0
1 0 0 01 0
Pair1->m 1 0 1 11 0
Pair2->1
Pair2->3
1 1 0 01 0
0 1 0 10 0
Pair2->m 0 1 1 01 1
Pairm->(m-1) 0 0 1 10 1
Table 2. Example of packet routing path table
-
8/2/2019 Reducing Total Power Consumption Method in Cloud Computing Environments
14/16
International Journal of Computer Networks & Communications (IJCNC) Vol.4, No.2, March 2012
82
5. Related work
Most of conventional measures to save the power consumed by servers and the network have
been discussed and implemented independently.A typical solution is to introduce power saving
technologies to ICT devices and data centers [8],[9]. Proposed measures to save the power
consumed by servers include not only introducing power saving technologies to them and using
a virtual device concept but also slowing processor clocks or reducing the number of operating
servers (turning the power of some servers off or putting them in sleep mode) while the total
load on the servers is small. Power saving technologies have been also actively developed in
the field of wireless LANs (IEEE802 wireless) and sensor networks [16]-[18], which
areessentially driven by batteries. These technologieshave been incorporated inwireless access
procedures. As for measures to save the power consumed by the network, References [10]-[14]
have proposed. Representative measures proposed include putting unused devices, lines or
nodes in sleep mode, controlling processor clock frequencies, controlling the line speed (rate
adaptation), the number of operating lines(link aggregation) or the number of active nodes
(route aggregation).
Moreover, the method to assign the power consumption by the network to an individual userhad not been fully discussed.
6. Conclusions
In order to reduce the total power consumption by cloud computing environments, this paper
first has identified the need of the collaboration among servers, communication network and
power network. Five fundamental policies for the collaboration and the algorithm to realize
each policy has been outlined. Next, this paper has proposed possible signaling sequences to
exchange information on power consumption between network and servers, in order to realize
the proposed policy. Then, in order to reduce the power consumption by the network, this paper
has proposed a method of estimating the volume of power consumption by all network devices
simply and assigning it to an individual user.
In the future, it is necessary to study the detailed algorithm for realizing each collaboration
policy and to clarifythe effectiveness of each policy on reducing total power consumption. It is
also necessary to study the optimal location of control functions for the signaling sequences toexchange information on power consumptionbetween network and servers. As for the
estimation of network power consumption, the impacts of imbalance between the volumes of
upstream and downstream traffic, multipoint connections and cases where multiple networks
areinvolved, are required to be studied.
Acknowledgment
We would like to thank Mr. Kenichi Hatakeyama for hisuseful suggestionson the signaling
sequences to exchange the informationon power consumption between network and servers.
References
[1] ITU Symposium on ICTs and Climate Change Summary Report, London, June 17&18,2008
http://www.itu.int/dms_pub/itu-t/oth/06/0F/T060F0060090001PDFE.pdf
-
8/2/2019 Reducing Total Power Consumption Method in Cloud Computing Environments
15/16
International Journal of Computer Networks & Communications (IJCNC) Vol.4, No.2, March 2012
83
[2] Green IT Initiative in Japan, METI, Japan Oct. 2008
http://www.meti.go.jp/english/policy/GreenITInitiativeInJapan.pdf
[3] J.W.Rittinghouse and J.F.Ransone: Cloud computing: Imprementation, management, and security,
CRC Press LLC, Aug. 2009.
[4] P.Mell and T.Grance, Effectively and securely using the cloud computing paradigm, NIST,Information Technology Lab., July 2009.
[5] P.Mell and T.Grance The NIST definition of cloud computing Version 15, 2009.
[6]S.Kuribayashi, Optimal Joint Multiple Resource Allocation Method for Cloud Computing
Environments, International Journal of Research and Reviews in Computer Science (IJRRCS),
Vol.2, No.1, pp.1-8, Feb. 2011
[7] S.Kuribayashi, Proposed congestion control method for cloud computing environments,
International journal of Computer Networks & Communications (IJCNC), Vol.3, No.5, pp.161-176,
Sep. 2011.
[8] M.Blackburn, Five ways to reduce data center server power consumption, Five ways to save
server power, the green grid. http://www.thegreengrid.org/
[9] VMware Distributed power Management (DPM) http://www.vmware.com/products/vi/vc/drs.html
[10] M.Gupta and S.Singh, Greening of the Internet, Proc.of ACM SIGCOMM03, pp.19-26, Aug.
2003.
[11]C.Gunaratne, K.Christensen, S.Suen, and B. Nordman, "Reducing the Power Consumption of
Ethernet with an Adaptive Link Rate (ALR)," IEEE Transactions on Computers, Vol. 57, No. 4, pp.
448-461, April 2008.
[12] F.Blanquicet, An power efficient Internet: some ongoing work, msrSeminar08 (June 2008).
[13] S.Nedevschi, L.Popa and G.Iannaccone, Reducing network power consumption via sleeping and
rate-adaptation, Proc. 5th USENIX Symposium on Networked Systems Design and
Implementation, April 2008.
[14] U.Lee, I.Rimac and V.Hilt, Greening the internet with content-centric networking, Proceedings ofthe 1st International Conference on Energy-Efficient Computing and Networking (2010).
[15] E.Jung and N.H.Vaidya, An power efficient MAC protocol for wireless LANs, Proc. IEEE
INFOCOM, June 2002.
[16] X.Wu, A.Jaekel and A.Bari, Optimal channel allocation with dynamic power control in cellular
networks, International Journal of Computer Networks & Communications (IJCNC) Vol.3, No.2,
March 2011
[17] B.Hohlt, L.Dohertly and E.Brewer, Flexible power scheduling for sensor networks, IEEE and
ACM Third International Symposium on Iformation Processing in Sensor Networks, April 2004.
[18] E.Jung and N.H.Vaidya, An Energy efficient MAC protocol for wireless LANs, In Proc. IEEE
INFOCOM, June 2002.
[19] F.Blanquicet and K.Christensen, Managing power use in a network with a new SNMP power stateMIB, IEEE Conference on Local Computer Networks (LCN) 2008, April 2008.
[20] Y.Matsumoto, S.Yanabu, A vision of an electric power architecture for the next generation,
Electrical Engineering in Japan Vol. 150, Issue 1 , pp. 18 25, January 2005.
[21] Power in Japan (2008),METI, Japan http://www.enecho.meti.go.jp/topics/power-in-
japan/english2008.pdf
-
8/2/2019 Reducing Total Power Consumption Method in Cloud Computing Environments
16/16
International Journal of Computer Networks & Communications (IJCNC) Vol.4, No.2, March 2012
84
[22] K.Hatakeyama, Y.Osana and S.Kuribayashi, Reducing total power consumption with collaboration
between network and servers, Proceeding of the 12-th International Conference on Network-Based
Information Systems (NBiS-2009), Aug. 2009.
[23]S.Kuribayashi, Reducing total ICT power consumption with collaboration among end systems,
communication network and power network, Proceeding of the 25th IEEE International Conferenceon Advanced Information Networking and Applications (AINA-2011), Mar. 2011
Author
Shin-ichiKuribayashireceived the B.E., M.E., and D.E. degrees from
Tohoku University, Japan, in 1978, 1980, and 1988 respectively. He joined
NTT Electrical Communications Labs in 1980. He has been engaged in the
design and development of DDX and ISDN packet switching, ATM, PHS,
and IMT 2000 and IP-VPN systems. He researched distributed
communication systems at Stanford University from December 1988 through
December 1989. He participated in international standardization on ATM
signaling and IMT2000 signaling protocols at ITU-T SG11 from 1990 through 2000. Since
April 2004, he has been a Professor in the Department of Computer and Information Science,Faculty of Science and Technology, Seikei University. His research interests include optimal
resource management, QoS control, traffic control for cloud computing environments and green
network. He is a member of IEEE, IEICE and IPSJ.